Specular reflections (i.e., highlight) always exist in endoscopic images, and they can severely disturb surgeons' observation and judgment. In an augmented reality (AR)-based surgery navigation system, the highlight may also lead to the failure of feature extraction or registration. In this paper, we propose an adaptive robust principal component analysis (Adaptive-RPCA) method to remove the specular reflections in endoscopic image sequences. It can iteratively optimize the sparse part parameter during RPCA decomposition. In this new approach, we first adaptively detect the highlight image based on pixels. With the proposed distance metric algorithm, it then automatically measures the similarity distance between the sparse result image and the detected highlight image. Finally, the low-rank and sparse results are obtained by enforcing the similarity distance between the two types of images to fall within a certain range. Our method has been verified by multiple different types of endoscopic image sequences in minimally invasive surgery (MIS). The experiments and clinical blind tests demonstrate that the new Adaptive-RPCA method can obtain the optimal sparse decomposition parameters directly and can generate robust highlight removal results. Compared with the state-of-the-art approaches, the proposed method not only achieves the better highlight removal results but also can adaptively process image sequences.

In recent years, an increasing number of medical engineering tasks, such as surgical navigation, pre-operative registration, and surgical robotics, rely on 3D reconstruction techniques. Self-supervised depth estimation has attracted interest in endoscopic scenarios because it does not require ground truth. Most existing methods depend on expanding the size of parameters to improve their performance. There, designing a lightweight self-supervised model that can obtain competitive results is a hot topic. We propose a lightweight network with a tight coupling of convolutional neural network (CNN) and Transformer for depth estimation. Unlike other methods that use CNN and Transformer to extract features separately and then fuse them on the deepest layer, we utilize the modules of CNN and Transformer to extract features at different scales in the encoder. This hierarchical structure leverages the advantages of CNN in texture perception and Transformer in shape extraction. In the same scale of feature extraction, the CNN is used to acquire local features while the Transformer encodes global information. Finally, we add multi-head attention modules to the pose network to improve the accuracy of predicted poses. Experiments demonstrate that our approach obtains comparable results while effectively compressing the model parameters on two datasets.